Theory of Low Frequency Contamination from Nonstationarity and Misspecification: Consequences for HAR Inference

Alessandro Casini; Pierre Perron; Taosong Deng

arxiv: 2103.01604 · v3 · submitted 2021-03-02 · 💰 econ.EM · math.ST· stat.TH

Theory of Low Frequency Contamination from Nonstationarity and Misspecification: Consequences for HAR Inference

Alessandro Casini , Taosong Deng , Pierre Perron This is my paper

Pith reviewed 2026-05-24 13:29 UTC · model grok-4.3

classification 💰 econ.EM math.STstat.TH

keywords low frequency contaminationnonstationarityHAR inferencelong-run varianceHAC estimatorsdouble kernel HACperiodogram bias

0 comments

The pith

Nonstationarity induces low frequency contamination that inflates long-run variance estimates and distorts HAR inference.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that general forms of nonstationarity create low frequency contamination in sample autocovariances and periodograms. This contamination produces asymptotic bias in long-run variance estimators, which in turn makes heteroskedasticity and autocorrelation robust tests undersized and prone to large power losses. The analysis distinguishes cases of finite-sample versus asymptotic effects and demonstrates that nonparametric smoothing over time avoids the contamination. It further separates the performance of different HAR procedures, with long bandwidths and fixed-b methods more vulnerable than standard HAC estimators.

Core claim

Nonstationarity produces explicit asymptotic bias in autocovariance and periodogram estimates through low frequency contamination; existing LRV estimators therefore tend to be inflated, HAR tests become undersized with dramatic power losses, long bandwidths or fixed-b procedures suffer more than HAC-based tests, and double kernel HAC estimators remain unaffected.

What carries the argument

Low frequency contamination in autocovariance and periodogram estimates induced by nonstationarity

Load-bearing premise

The derivations assume a general form of nonstationarity that induces the described low frequency contamination in autocovariance and periodogram estimates.

What would settle it

A Monte Carlo experiment or empirical series known to contain low-frequency nonstationarity in which standard HAC LRV estimates exceed the true long-run variance while double kernel HAC estimates remain close to it.

read the original abstract

We establish theoretical results about the low frequency contamination (i.e., long memory effects) induced by general nonstationarity for estimates such as the sample autocovariance and the periodogram, and deduce consequences for heteroskedasticity and autocorrelation robust (HAR) inference. We present explicit expressions for the asymptotic bias of these estimates. We distinguish cases where this contamination only occurs as a small-sample problem and cases where the contamination continues to hold asymptotically. We show theoretically that nonparametric smoothing over time is robust to low frequency contamination. Our results provide new insights on the debate between consistent versus inconsistent long-run variance (LRV) estimation. Existing LRV estimators tend to be in inflated when the data are nonstationary. This results in HAR tests that can be undersized and exhibit dramatic power losses. Our theory indicates that long bandwidths or fixed-b HAR tests suffer more from low frequency contamination relative to HAR tests based on HAC estimators, whereas recently introduced double kernel HAC estimators do not super from this problem. Finally, we present second-order Edgeworth expansions under nonstationarity about the distribution of HAC and DK-HAC estimators and about the corresponding t-test in the linear regression model.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper derives explicit asymptotic bias formulas for autocovariances and periodograms under general nonstationarity and shows double-kernel HAC avoids the resulting contamination that hits other HAR methods.

read the letter

The central contribution is the set of explicit bias expressions for the sample autocovariance and periodogram when the series has a general nonstationary component. These formulas let the authors separate cases where the low-frequency contamination is only a small-sample issue from cases where it persists in the limit. They then use the same setup to compare HAR procedures: long-bandwidth and fixed-b tests suffer more, while the double-kernel HAC estimators recently proposed in the literature remain unaffected. The second-order Edgeworth expansions for the estimators and the associated t-statistics in the linear model are a useful addition for understanding size and power under the same conditions. This gives a cleaner account of why inconsistent LRV estimators can produce undersized tests and power losses when trends or breaks are present. The derivations appear self-contained and the distinction between estimator classes is new relative to the cited work on HAR. The main limitation is that the nonstationarity class is stated at a high level in the abstract; without the precise regularity conditions it is hard to judge how broad the robustness claim really is. The power-loss language is also strong for a purely theoretical paper, though the bias results themselves do not require simulation support. The work is aimed at specialists in econometric time-series methods who already use or develop HAC and HAR procedures. It is technically grounded enough to merit referee time even if revisions are needed on the scope of the assumptions.

Referee Report

0 major / 3 minor

Summary. The paper establishes theoretical results on low frequency contamination (long memory effects) induced by general nonstationarity in sample autocovariance and periodogram estimates, providing explicit asymptotic bias expressions. It distinguishes small-sample versus asymptotic contamination cases, shows robustness of nonparametric smoothing over time, derives consequences for HAR inference (including that existing LRV estimators inflate under nonstationarity, leading to undersized tests with power losses, with long bandwidth/fixed-b tests suffering more than standard HAC while double-kernel HAC estimators are unaffected), and presents second-order Edgeworth expansions under nonstationarity for HAC/DK-HAC estimators and associated t-tests in linear regression.

Significance. If the derivations hold, the work supplies explicit bias formulas and a robustness ranking among LRV estimators that directly informs the consistent-versus-inconsistent LRV debate in econometrics. The Edgeworth expansions and the distinction between small-sample and asymptotic contamination constitute concrete, falsifiable contributions that can guide finite-sample practice and estimator choice when low-frequency nonstationarity is plausible.

minor comments (3)

Abstract: 'do not super from this problem' is a typographical error and should read 'do not suffer from this problem'.
The manuscript would benefit from an explicit statement, early in the introduction or §2, of the precise class of nonstationary processes for which the bias expressions are derived (e.g., the functional form of the time-varying spectrum or the rate at which low-frequency power accumulates).
Notation for the double-kernel HAC estimator should be introduced once and used consistently; the current text alternates between 'DK-HAC' and 'double kernel HAC' without a single defining equation.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. No specific major comments were raised in the report, so we have no point-by-point responses to provide at this stage. We will incorporate any minor editorial suggestions in the revised version.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper derives explicit asymptotic bias expressions for sample autocovariance and periodogram under a stated general nonstationarity setup, then deduces consequences for HAR inference including robustness distinctions for double-kernel HAC. These steps are presented as first-principles results from the nonstationarity assumptions rather than reductions to fitted parameters, self-definitions, or load-bearing self-citations. The abstract and description indicate self-contained theoretical content with no quoted step that equates a claimed prediction or uniqueness result to its own inputs by construction. This is the normal case of an independent derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard mathematical assumptions for asymptotic analysis in time series and the specific nonstationarity model used to derive contamination effects.

axioms (1)

domain assumption General nonstationarity in the time series process induces low frequency contamination in autocovariance and periodogram estimates
The paper relies on assumptions about the form of nonstationarity to derive the bias expressions and distinguish small-sample vs asymptotic cases.

pith-pipeline@v0.9.0 · 5746 in / 1314 out tokens · 29317 ms · 2026-05-24T13:29:21.006523+00:00 · methodology

Theory of Low Frequency Contamination from Nonstationarity and Misspecification: Consequences for HAR Inference

Core claim

What carries the argument

Load-bearing premise

What would settle it

discussion (0)